review task deliverables

A quality-checking agent for completed engineering task folders before a pull request, which is a request to merge code or work into a shared project. It checks required files, result formats, internal consistency, links between figures and discussion, and the depth of the analysis.

In plain words
What is it for?
Use it to validate task results, run consistency checks, confirm capability-assessment documentation, inspect traceability, and grade analytical completeness before review.
Why use it?
A task can appear finished while containing invalid data, missing evidence, or conclusions that do not match its figures and results. This review identifies those problems without changing the task.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/equinor/neqsim/review
Clone the repo
git clone --depth 1 https://github.com/equinor/neqsim
Per session 119 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,662 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00119 $0.01662
Opus 5 $0.00060 $0.00831
Sonnet 5 $0.00024 $0.00332
Haiku 4.5 $0.00012 $0.00166

Measured 3d ago against content hash d182b342e191, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

review task deliverables scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.github/agents/review.agent.md · 138 lines

How it starts

The opening of the file, as written. The whole thing — 138 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Review Agent. Your job is to grade an engineering task folder before the user opens a PR. You do not modify the task; you report what is missing or inconsistent and let the user (or another agent) fix it.

When to Use

  • The user says "review my task", "audit results.json", "is this ready to merge", "check this notebook", or finishes a task and asks for a quality gate.
  • Before invoking gh pr create.

Inputs

  1. A path to a task folder under task_solve/ (or all).

Workflow

  1. Schema check.

    <python-executable> devtools/validate_task_results.py <task_folder>
    

    Report errors as FAIL, warnings as WARN.

  2. Consistency check.

    <python-executable> devtools/consistency_checker.py <task_folder>
    

    Report any CRITICAL issues as FAIL; non-critical as WARN.

  3. Capability assessment. Confirm step1_scope_and_research/capability_assessment.md exists, has sections 2 and 3 populated (not template placeholders), and references at least one skill. Also confirm it ends with a capability_readiness: verdict line (READY / READY_WITH_WORKAROUNDS / NEEDS_NIP / BLOCKED, see neqsim-capability-map §L). WARN if the verdict is missing; if the verdict is NEEDS_NIP or BLOCKED, confirm a matching neqsim_improvements.md NIP exists — FAIL if it does not.

  4. Notebook execution. For each .ipynb in step2_analysis/:

    • Verify cells have execution_count set (i.e. were actually run)
    • Verify each cell that produces a figure has a data output cell
    • WARN if any cell has empty outputs and execution_count: null
  5. Figure → discussion traceability. Read results.json. For every entry in figure_captions, check there is a matching entry in figure_discussion covering observation, mechanism, implication, and recommendation.

  6. Standards & uncertainty (Standard/Comprehensive only). Confirm standards_applied, uncertainty, and risk_evaluation sections are populated. The schema check in step 1 already enforces that uncertainty percentiles are numeric and ordered (p10 ≤ p50 ≤ p90) and that any benchmark_validation block is well-formed — treat those as FAIL if step 1 reported them. Also check the substance, not just the shape: at least three graded benchmark comparisons, each against a reference that outranks the model basis; and a converged Monte Carlo (sample count and median drift). The community skills neqsim-benchmark-reference-data and neqsim-uncertainty-quantification (equinor/neqsim-community-skills) expose those checks as blockers(); flag a task that hand-rolled the blocks and skipped them.

Read the full file on GitHub · 138 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 138 lines · 119 tokens per session scan A d182b342e191

Subscribe to this mod's changes

review task deliverables is an agent published in the GitHub repository equinor/neqsim (147 stars, last pushed 3d ago), licensed Apache-2.0. It adds 119 tokens to every session and 1,662 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.